ANALYSIS OF FEATURE REDUCTION TECHNIQUES FOR HUMAN INTERACTION DETECTION

This research paper elaborates on how different feature reduction processes are applied on human joint coordinate data, used to train a machine learning model for detection and classification of eight distinct interactions between two people with high accuracy and low computation time. We believe that a low-complexity machine learning algorithm such as SVM, when combined with the appropriate feature reduction technique can help reduce the trade-off between model accuracy and computation time. Two feature reduction methods have been explored in detail, namely Principal Component Analysis and Univariate Feature Selection. We have recorded our observations of the positive effect of these methods on the model, which uses joint motion and joint distance features for both real time detection and full sequence classification.